Given intrinsic uncertainty in initial condition (IC) used by NWP models, Lorenz pointed out the predictability issue of atmospheric motion. The most important application of this predictability theory might be the ensemble forecasting. Recent years, as ensemble approach becomes widely used in both operational and research communities, more and more scientists paid their attention to the model physics uncertainty in constructing an ensemble prediction system (EPS).

A theoretical question raised is that the predictability issue is mainly an initial value problem or model physics problem or both? The practical aspect of this question is how much extra value or useful uncertainty information will be gained by adding model physics diversity to an EPS since maintainance and development of a variety of valid physics schemes is difficult, complex and costy. In other words, the cost/benefit ratio is a very real issue in operational world. Therefore, the study of this issue could be extremely important.

Two experiments are designed to study the above mentioned issue. One is the comparison between a solely IC-based ensemble and a solely different physics-based ensemble where the NCEP Eta model is used. Another is to compare the NCEP physics-diversity SREF system with the same system but with some extra physics options turned off. Results from a couple of case studies will be presented.